Updated daily503 verified stories
Safety News tracker.
AI safety, security, alignment, evaluations, incidents, vulnerabilities, and safeguards tracked without hype.
- 54stories this week
- 503sources cited
- 503verified timeline entries
Verified sourcing
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
Plain English
What happened, why it matters, and what to watch — without the jargon.
No filler
When the signal is thin, we publish nothing rather than padding the feed.
More stories
9 storiesSecurity
Study finds longer LLM oversight reviews become more rejective, not more discriminative
An arXiv study reports that reviewing longer action sequences increases both error detection and false rejection, with the best balance among six tested LLM judges occurring at one or two actions.arxiv.orgSecurity
Gartner survey ranks AI vulnerability discovery as highest-impact emerging risk
Help Net Security reports that a Gartner survey of 316 companies ranked AI-driven vulnerability discovery as the most impactful of 20 emerging risks, while respondents also rated themselves highly prepared for it.helpnetsecurity.comSecurity
NYT: Zhipu AI’s planned GLM 5.3 release renews open-weight cybersecurity debate
The New York Times reports that China’s Zhipu AI plans to release GLM 5.3 as open-weight software, intensifying debate over whether broadly accessible AI systems create greater cyber risk or expand defensive capabilities.cn.nytimes.comSecurity
BBC reports AI-generated fake-doctor videos spreading among older viewers in Taiwan
BBC News Chinese reports that AI-generated videos featuring fictional doctors and false health or welfare claims are circulating widely among older people in Taiwan. A Taipei class is teaching seniors to question videos shared through YouTube and messaging groups, while the scale, ownership and possible political…bbc.comSecurity
SyPS measures how prompt framing changes sycophancy in language models
A paper introduces SyPS, a framework for testing whether changes in confidence, emotion, social consensus and validation-seeking language make large language models more likely to agree with users.arxiv.orgSecurity
BigGo Finance reports field test finds AI vulnerability research works best with human guidance
A months-long field test presented by web-security researcher James Kettle found that AI struggled to invent novel attack paths independently but became highly productive when guided by an experienced human. BigGo Finance reports that the work produced a previously unstudied vulnerability domain, though the…finance.biggo.comSecurity
Position paper argues AI agents can erode the human oversight they rely on
A new arXiv position paper argues that giving AI agents more autonomy may weaken the human judgment needed to supervise them effectively.arxiv.orgSecurity
ChosunBiz reports Microsoft security executive says multi-agent AI improves vulnerability detection
ChosunBiz reports that Microsoft corporate vice president Kim Tae-su said multi-agent workflows can find and verify software vulnerabilities more efficiently than a single AI system. The report describes results from DARPA’s AI Cyber Challenge and Microsoft’s MDASH system, but the claims have not been independently…biz.chosun.comSecurity
Herald Business reports South Korea seeks domestic cybersecurity AI model
The Herald Business reports that South Korea is accepting applications for a government-backed project to develop a domestic AI foundation model focused on cybersecurity. The selected team would receive 256 B200 GPUs for 10 months, although the project, applicants and expected capabilities have not been…biz.heraldcorp.com
One useful briefing each week
Keep up with AI without living in the feed.
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
Reach people who are learning AI
Hiring an AI professional or launching a useful AI product? Put it in front of people who came here to learn and act.
Post an AI jobSubmit an AI tool